[CMake] Why CPack does not _ALWAYS_ use DESTDIR?

[Alexander Neundorf]

On Tuesday 13 May 2008, Bill Hoffman wrote:
> Alexander Neundorf wrote:
> > On Tuesday 13 May 2008, Bill Hoffman wrote:
> >> Eric Noulard wrote:
> >
> > ...
> >
> >>> Using absolute path install on unix is not a good idea either but
> >>> you sometimes have to use it to put you file in /etc/ .
> >>>
> >>> Be sure I'm not willing to break some backward compatibilty but
> >>> to avoid strange behavior when building installer.
> >>
> >> There is one more issue.  CPack is supposed to work with any build
> >> system not just CMake.
> >
> > Is this an important feature ? I mean, cpack is not really wide spread
> > yet, and I think it mostly has advantages if it's integrated with the
> > rest of the build done by cmake. At least requiring working DESTDIR
> > support would be acceptable IMO (I don't think compatibilty requirements
> > for cpack are already as strong as for cmake, since I guess much less
> > people are already using it. IMO cpack is just about to become mature.).
>
> We have projects that use it for non-cmake parts of the build.  I don't
> want to break those projects.  In fact, the project that funded CPack is
> the one I am most concerned about breaking.
>
> > I actually found Erics argument that the default behaviour is not working
> > for absolute install dirs with any package generator quite good.
> >
> > The rpm and deb generators are new, would it be ok to enable DESTDIR for
> > these two unconditionally ? There can be nobody who relies on this
> > behaviour for these two package types yet.
>
> I really don't like the idea of different defaults for different
> generators.  Hey, my stuff worked here, but not here type of stuff.
> IMO, the use of absolute dirs is a bug. :)

We support it in KDE, since packagers want/need it for the different 
components KDE installs. So, I don't think they agree it's a bug ;-)

Alex